Source Attribution of Cyanides Using Anionic Impurity Profiling

Chemical attribution signatures (CAS) for chemical threat agents (CTAs), such as cyanides, are being investigated to provide an evidentiary link betwe...
0 downloads 0 Views 695KB Size
Subscriber access provided by University of Birmingham

Article

Source Attribution of Cyanides using Anionic Impurity Profiling, Stable Isotope Ratios, Trace Elemental Analysis and Chemometrics Nikhil S. Mirjankar, Carlos G. Fraga, April J. Carman, and James J. Moran Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.5b04126 • Publication Date (Web): 28 Dec 2015 Downloaded from http://pubs.acs.org on January 5, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

PNNL-SA-113038

Source Attribution of Cyanides using Anionic Impurity Profiling, Stable Isotope Ratios, Trace Elemental Analysis and Chemometrics

Nikhil S. Mirjankar, Carlos G. Fraga*, April J. Carman, and James J. Moran

Pacific Northwest National Laboratory 902 Battelle Boulevard Richland, Washington 99352, USA *E-mail: [email protected]

Manuscript Submitted to Analytical Chemistry

1 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 25

Abstract Chemical attribution signatures (CAS) for chemical threat agents (CTAs), such as cyanides, are being investigated to provide an evidentiary link between CTAs and specific sources to support criminal investigations and prosecutions. Herein, stocks of KCN and NaCN were analyzed for trace anions by high performance ion chromatography (HPIC), carbon stable isotope ratio (δ13C) by isotope ratio mass spectrometry (IRMS) and trace elements by inductively coupled plasma optical emission spectroscopy (ICP-OES). The collected analytical data were evaluated using hierarchical cluster analysis (HCA), Fisher-ratio (F-ratio), interval partial least squares (iPLS), genetic algorithm-based partial least squares (GAPLS), partial least squares discriminate analysis (PLSDA), K nearest neighbors (KNN), and support vector machines discriminate analysis (SVMDA). HCA of anion impurity profiles from multiple cyanide stocks from six reported countries of origin resulted in cyanide samples clustering into three groups, independent of the associated alkali metal (K or Na). The three groups were independently corroborated by HCA of cyanide elemental profiles and corresponded to countries each having one known solid cyanide factory: Czech Republic, Germany, and United States. Carbon stable isotope measurements resulted in two clusters: Germany and United States (the single Czech stock grouped with United States stocks). Classification errors for two validation studies using anion impurity profiles collected over five years on different instruments were as low as zero for KNN and SVMDA, demonstrating the excellent reliability associated with using anion impurities for matching a cyanide sample to its factory using our current cyanide stocks. Variable selection methods reduced errors for those classification methods having errors greater than zero; iPLS-forward selection and F-ratio typically provided the lowest errors. Finally, using anion profiles to classify cyanides to a specific stock or stock group for a subset of United States stocks resulted in cross-validation errors ranging from zero to 5.3%.

2 ACS Paragon Plus Environment

Page 3 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Introduction Solid cyanides are chemical threat agents (CTAs) because they are highly toxic, stable, and readily available. Cyanides have been identified by the U.S. Centers for Disease Control and Prevention as one of the most probable agents for chemical terrorism.1 Furthermore, recurring incidents involving the use of cyanides in numerous high profile crimes, including recent ones, have made them a highly publicized and notorious poison.2-4 Chemical attribution signatures, such as impurities and other chemical and physical characteristics of a CTA, may provide key evidentiary links between events such as these and a specific source of cyanide (e.g., a suspect’s cyanide supply). The toxicity of cyanides arises from their high affinity for transition metals.5 In mammalian cells, this promotes binding to the iron in heme groups which disrupts aerobic respiration leading to cellular asphyxiation and death.6 Despite their toxicity, cyanides see widespread use in precious metal mining (due to their strong chelation of valuable metals) with additional industrial applications including electroplating and surface treatment, and in scientific research.7 To meet these demands, commercial sodium and potassium cyanide (NaCN and KCN respectively) are produced in large quantities but by relatively few commercial manufacturers. Typical synthesis utilizes neutralization of hydrogen cyanide (or hydrocyanic acid) with aqueous solutions of sodium hydroxide or potassium hydroxide, to precipitate solid phase NaCN and KCN respectively.7 While the general synthesis steps for cyanides are nearly uniform across different manufacturers, various factors including the reagents, manufacturing process, and local geology can each influence impurities and stable isotope signatures that remain in the final product, and in doing so impart a chemical attribution signature (CAS). The potential forensic merit of organic and inorganic impurities as CAS for CTAs has been previously demonstrated.8-13 Previous work using 8 KCN stocks reportedly from four countries suggested that CAS might be used to determine country (or manufacturing facility) of origin.9 In this study, sample resolution was made possible by the analysis of anion impurities in KCN stocks by high performance ion chromatography (HPIC) with anion-exchange chromatography and conductivity detection and by applying chemometric analysis for pattern recognition and classification. The current work leverages methods developed in the previous 3 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 25

investigation and explores source attribution in a larger set of cyanide stocks, including for the first time NaCN stocks. The current study further employs and integrates data from two complementary analytical approaches known to provide CAS; isotope ratio mass spectrometry (IRMS) and inductively coupled optical emission spectroscopy (ICP-OES). Stable carbon and stable nitrogen isotope ratios from IRMS have been reported to be useful forensic signatures for sample matching of KCN and NaCN stocks, however, the relationship between cyanide stable isotope ratios and manufacturer has not been addressed.14,15 Elemental profiling of cyanides by ICP-OES and similar techniques has not been reported as potential CAS for cyanides. Yet, ICP-MS has provided elemental profiling which, when combined with isotopic analysis, has recently been demonstrated as CAS for a very similar system - discrimination of ammonium nitrate-based fertilizers based on manufacturer and stock.16 Both cyanide and ammonium nitrate are inorganic salts made in aqueous environments. The general similarities in molecular form and manufacturing between ammonium nitrate and cyanide permit the leveraging of independent research projects for the overall advancement of chemical forensics, i.e., the scientific discipline underpinning the source attribution of chemicals and chemical mixtures. Herein, for the first time, we describe discrimination of commercial cyanide samples by presumed factory using an integrated anionic, elemental, and isotopic profiling approach. Further, we demonstrate the ability of impurity profiling to match specific cyanide stocks. CAS robustness was also studied using samples prepared and analyzed over a five year period with different operators and different instrument models. Finally, several chemometric variable selection methods and supervised classification methods were evaluated for classification performance using anion impurity data. Experimental Cyanides The commercial cyanides used in this study included 13 solid KCN and 14 solid NaCN stocks with reported purities of at least 95.0%. Each stock was identified by a supplier, product number, and unique lot number as listed on the original supplier container. The country of origin for each stock was obtained through a certified 4 ACS Paragon Plus Environment

Page 5 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

document or e-mail message from the supplier. In addition, the supplier provided the year each stock was manufactured, acquired or analyzed for quality control. Table 1 columns 1 – 6 list relevant stock information for all 27 cyanide stocks. HPIC-1, HPIC-2, and HPIC-3 Analysis The HPIC analysis of 27 cyanide stocks was conducted in three separate batches. The first analysis batch (HPIC-1) involved eight KCN stocks denoted in Table 1 column 7 (HPIC-1 8/2009). The second HPIC batch (HPIC-2) involved 12 NaCN stocks and two KCN stocks denoted in Table 1 column 8 (HPIC-2 9/2011). The third HPIC batch (HPIC-3) involved six KCN and four NaCN stocks denoted in Table 1 column 9 (HPIC-3 1/2015). Sample preparation and analysis details for each HPIC batch are provided in the Supporting Information. Very briefly, each HPIC batch included the dissolution of 0.1 to 1.0 g aliquots of cyanide stocks in18 MΩ-cm deionized water to create 1000 ppm (w/w) aqueous cyanide samples that were analyzed using an HPIC system fitted with an anion-exchange column and conductivity detector. Thirteen anions were identified in the cyanide samples using the retention times of anion standards as references. The HPIC-1 data were originally collected and used in a previous study.9 IRMS-1 and IRMS-2 Analysis Carbon and nitrogen stable isotope analyses (δ13C and δ15N) of 27 cyanide stocks were conducted in two batches (Table 1: IRMS-1 and IRMS-2) using an elemental analyzer (EA) coupled to an isotope ratio mass spectrometer (IRMS). The experimental details for each IRMS batch are described in the Supporting Information. Very briefly, 0.3 to 1.4 mg aliquots of cyanides stocks were sealed into tin capsules and then converted by the EA to CO2 and N2 for downstream stable isotope analysis by IRMS. All isotope values were reported in delta (δ) notation where δ= (RSa/RStd – 1)*1000 ‰, RSa and RStd equal the isotope ratio (13C/12C or 15

N/14N) of the sample and an internationally recognized isotope standard respectively. The IRMS-1 data were

originally collected and used in a previous study.15 ICP-OES-1 and ICP-OES-2 Analysis 5 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 25

The ICP-OES analysis of 20 cyanide stocks was conducted in two batches (Table 1: ICP-OES-1 and ICP-OES2) for nine elements: Ba, Ca, Fe, Mg, P, Rb, S, Si, and Sr. The experimental details for each ICP-OES batch are described in the Supporting Information. Chemometric Analysis Chemometric analysis was implemented using PLS Toolbox 7.9 (Eigenvector Research Inc., Manson, WA) with Matlab R2014a (Mathworks Inc., Natick, MA). Initially the data sets were investigated with hierarchical cluster analysis (HCA) and principal component analysis (PCA). HCA and PCA are unsupervised techniques often used for pattern recognition and exploratory data analysis to probe for evidence of sample clustering. Partial least squares discriminant analysis (PLSDA) was utilized for supervised classification modeling of the impurity profile data sets. PLSDA is useful for directly identifying variations in data space that discriminate classes with less noise compared to linear discriminant analysis (LDA) and with variable selection advantages of PLS regression modeling.17 The data were additionally analyzed by two other commonly used supervised classification methods, K nearest neighbor (KNN) and support vector machines discriminate analysis (SVMDA). Classification performance for a given supervised method (KNN, PLSDA or SVMDA) was measured by the classification error for a set of samples in a validation set that were completely separate from a set of training samples used in creating a classification model. Classification error was the ratio of the number of misclassified profiles to the total number of profiles in the validation set. In the absence of a validation set, error for a classification model was measured as the average cross-validation error (E) obtained for all classes in the training set.18 E equaled 1 – NE, where NE was the non-error fraction (0 - 1) calculated as the average of class sensitivity and selectivity for all classes where the sensitivity and selectivity for each class were calculated as follows: class sensitivity = TP/(TP + FN) class selectivity = TN/(FP+TN) 6 ACS Paragon Plus Environment

Page 7 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

where TP is true positive, TN is true negative, FP is false positive, and FN is false negative. PLS Toolbox automatically calculated E using cross-validation. For both validation and cross-validation classification, each test sample was matched to its probable class which meant every test sample had to be assigned to a modeled class. Variable selection was performed using three different approaches, Fisher- ratio (F-ratio) method, interval PLS (iPLS), and genetic algorithm-based PLS (GAPLS), in order to retain variables that have discriminatory information correlated with known sample classes and discard noisy or irrelevant variables. Brief explanations and helpful references on the selected variable selection methods are provided in the Supporting Information. Results and Discussions HPIC, IRMS and ICP-OES data A total of 120 anion HPIC analyses of aqueous cyanide samples prepared from 27 solid cyanide stocks were performed in three batches (HPIC-1, HPIC-2, and HPIC-3) over a period spanning more than five years (see Table 1). A total of 13 anions were identified in the cyanide samples using the retention times of anion standards as references. The anions that were specifically identified were chloride (Cl), nitrite (NO2), carbonate (CO3), bromide (Br), sulfate (SO4), nitrate (NO3), oxalate (OX), and phosphate (PO4). Five anions that were not specifically identified by chemical name were labeled sequentially unk1 through unk5. Figure 1 depicts two representative anion chromatograms for a KCN stock and a NaCN stock reportedly from Germany. These two chromatograms from HPIC-3 were selected to primarily illustrate the relative levels of anions in the cyanide stocks; the concentrations for five anions are also listed in Figure 1 as a reference for absolute anion levels in cyanides. Eleven of the denoted anions (unk1, unk2, Cl, unk3, NO2, unk4, CO3, unk5, SO4, OX, and PO4) were selected to represent each sample’s anion profile; bromide and nitrate were not included because they were originally excluded in HPIC-1.9 For consistency, peak areas were used for all anion profiles owing to the inclusion of unknown anions (unk1- unk5), which lacked standardized concentration measurements. The peak areas per analysis for the 11 anions are provided in Table S-1 for all HPIC cyanide analyses. In terms of IRMS 7 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 25

data, Table S-2 lists the measured stable carbon and nitrogen isotope ratios for 27 cyanide stocks from IRMS-1 and IRMS-2. In terms of ICP-OES data, Table S-3 lists the concentrations for nine elements in samples from 20 cyanide stocks measured in two batches (ICP-OES-1 and ICP-OES-2). Cluster Analysis HCA was carried out to identify patterns in the HPIC, IRMS, and ICP-OES data for the cyanide samples. Previous HCA demonstrated that KCN samples tightly clustered according to reported country of origin using the concentrations of four anion impurities in cyanides: unk5, SO4, OX, and PO4.9 Using the peak areas for the same four anions from the combined HPIC data (HPIC-1, HPIC-2, and HPIC-3) resulted in the cyanide samples clustering into three main groups as shown in Figure 2. The three groups, designated by countries of origins, were (1) GE/BE, (2) CZ, and (3) US/UK/ES. Unlike the previous work on KCN, there were 10 outlier samples (see Figure 2) whose 4-anion profiles caused them to group outside other samples from the same group. All replicate samples (a,b,c) from cyanide stocks K-US-6, N-GE-3, and N-BE-3 were considered outliers; K-US-6 sample profiles grouped with the GE/BE samples and not with profiles from the other nine US cyanide stocks while sample profiles from N-GE-3 and N-BE-3 grouped with the US/UK/ES samples and not with those from nine other German and Belgium stocks. These outliers are explained in the interpretation section of this paper. One sample profile N-US-4c did not cluster with the other two replicate profiles (N-US-4a and N-US-4b) and can be explained as an analysis fault. N-US-4c was removed from subsequent data analyses. Another observation, obscured by the fixed scale for Figure 2 but apparent in a zoomed-in view (not shown), is that some samples form sub-clusters according to stock. These sub-clusters are later addressed in the classification section of this paper. Any sub-clustering based on cyanide type (NaCN or KCN) or analysis batch was not observed. HCA on the averaged stable nitrogen isotope ratios from IRMS-1 (see Table S-2) showed no obvious clustering for cyanides based on cyanide type or country-of-origin. In terms of averaged stable carbon isotope ratios, cyanide stocks mostly clustered according to country of origin as depicted by the HCA dendrogram in Figure 3. While HCA is considered a multivariate technique, it also works for revealing sample clusters in univariate data 8 ACS Paragon Plus Environment

Page 9 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

like carbon isotope ratios. As shown in Figure 3, the cyanide stocks cluster into two groups, US/UK/ES and GE/BE, the same groups obtained with the HPIC anion data minus the CZ group. The CZ stock falls within the US/UK/ES group based on its carbon isotope ratio, in contrast to its independent clustering based on HPIC data. As with the HPIC anion data, samples from the outlier US stock (K-US-6) cluster with the GE/BE group and samples from the other two outlier stocks (N-GE-3 and N-BE-3) cluster with the US/UK/ES group based on the carbon IRMS data. Interestingly, as shown in Figure 3, the N-US-4 stock clusters with the GE/BE group; that was not the case with the HPIC data. A highly probable reason for that is discussed in the interpretation section of this paper. HCA on the combined ICP-OES data (ICP-OES-1 and ICP-OES-2) resulted in cyanide samples clustering into three main groups designated by countries of origins: (1) GE/BE, (2) CZ, and (3) US/UK (see Figure 4). These are the same groups observed for HPIC data except ES was not in a group because stocks reportedly from Spain were acquired after the ICP-OES analyses and therefore not analyzed. As in the case with the HPIC data, most samples clustered with other samples from the same country of origin or same country group. The only exceptions were samples from stocks N-BE-3 and N-GE-3 that grouped with the US/UK samples as was the case for the HPIC and IRMS data sets. In terms of any clustering according to analysis batch or cyanide type, the ICP-OES data did not lend itself to that assessment because one analysis batch involved KCN stocks while the other involved NaCN stocks; therefore their effects, if any, were not separable. Finally, it does appear in Figure 4 that KCN samples sub-cluster according to stock indicating measurable differences among KCN stocks from the same group that are reproducible among different samples from the same stock; this could not be determined for the NaCN stocks because unlike the KCN stocks, only one solid sample was taken from each NaCN stock. Interpretation As shown by HCA on the combined HPIC data (Figure 2) and combined ICP-OES data (Figure 4), cyanide stocks cluster into three groups: (1) GE/BE, (2) CZ, and (3) US/UK/ES. These three groups corresponded to the three countries where the cyanide stocks in this study were apparently manufactured: Germany, the Czech 9 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 25

Republic and the United States. This is supported by known cyanide factory locations and operations. For instance, each listed country group contains only one country that is an actual producer of solid NaCN and solid KCN, i.e., Germany, the Czech Republic, and the United States.19,20 Also, while Belgium and the United Kingdom are cyanide producers, their cyanide production includes only NaCN solutions and not KCN. Solid cyanides are distributed from factories as solids and not as solutions that are later evaporated and redistributed as solids19; hence, the United Kingdom and Belgium cannot possibly be countries of manufacture for the solid NaCN and solid KCN in this study. The only cyanide factory in Spain produced NaCN solutions and stopped production in 200319; hence, Spain cannot be a country of manufacture for the solid cyanides in this study. Given this information, it is highly likely that the solid cyanide stocks used in this study were manufactured in only three countries: Germany, the Czech Republic and the United States. Implicit in this understanding is that the country of origin listed by a manufacturer may differ from the location the bulk cyanide was manufactured. In many cases it is not possible or required for suppliers to verify the original source of chemicals they are provided and in some cases the term country-of-origin21 may be interpreted differently among organizations. Based on our work, the country of origin reported by suppliers did not match the believed country of manufacture in 12 out of 27 stocks or ~ 44%. The 12 stocks included the nine stocks reportedly from Belgium, Spain, and the United Kingdom and the three outlier stocks N-GE-3, N-BE-3, and K-US-6 whose samples did not group with other stock samples from the same reported countries of origins. The IRMS data supports the conclusion that the cyanide stocks from the GE/BE group were manufactured in Germany and the stocks from the US/UK/ES group were manufactured in the United States. Figure 3 shows cyanides stocks clustering into two groups encompassing a United States and German source. As with the HPIC data, the reported Belgium stocks grouped with the German stocks and the reported United Kingdom and Spanish stocks grouped with the US stocks. The HPIC (Figure 2), IRMS (Figure 3), and ICP-OES (Figure 4) data for the outlier samples strongly suggests that N-GE-3 and N-BE-3 were made in the US, and K-US-6 was made in Germany, contrary to the reported countries of origin.

10 ACS Paragon Plus Environment

Page 11 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Another outlier was stock N-US-4 whose HPIC and ICP-OES data pointed to the US as a source but whose IRMS data suggested a German origin. This apparent contradiction is potentially explained by the fact the NUS-4 was manufactured no later than 1974 while the other cyanides in this study were likely manufactured approximately between 2003 and 2012 (see Table 1). We therefore postulate that the carbon source and/or process utilized to make the HCN used in cyanide production changed between the 1970’s and 2000’s, creating a shift in the associated stable carbon isotope signature for N-US-4 relative to the other US stocks. The source of the trace anions and elemental impurities in the cyanides is presumably the water used to make the alkalihydroxide reagent used to neutralize the HCN. As the elemental content and anion data for N-US-4 is consistent with more modern US cyanide stocks, the water source used in manufacturing the US samples appears to have remained mostly unchanged. The dependence of HPIC and ICP-OES data on the water used in manufacturing is also the reason why cyanides regardless of type (KCN or NaCN) group together based on country of origin. Given that each country has only a single manufacturing facility for solid cyanides;19,20 it seems reasonable that samples from that facility would be manufactured using a consistent, common water source. This hypothesis is reinforced by a cyanide manufacturer who stated that the KOH and NaOH solutions used in their facility originate from the same producer who likely uses local brine and local water to make the aqueous hydroxide solutions.22 Lastly, isotopic analysis grouped (Figure 3), the K-CZ-D stock with the US stocks in contrast to its separation by the HPIC and ICP-OES data. This illustrates that even two countries that are geographically far apart from one another may have similar or indistinguishable chemical signatures in their chemical products; likely resulting from similar input signatures leaving similar CAS in the final cyanide products. Similarly, nitrogen isotope analysis provided minimal sample resolution, most likely since atmospheric N is both the precursor for cyanide formation and is well mixed with very little isotopic variability around the planet. The comparisons between the K-CZ-D stocks and US/UK/ES stocks exemplify why accurate sample discrimination benefits from compilation of multiple data types, because, a single data stream may contain overlapping signals between unrelated samples. 11 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 25

Supervised Classification by Factory As discussed above, we have deduced that there are only three countries of manufacture for solid cyanides (in this study) and each one is a specific factory. Based on that deduction, we conducted classification studies to determine how well cyanide samples can be classified according factory using different combinations of chemometric variable selection methods and classification methods. Such an effort would give an indication to how well unknown cyanide samples can be matched to their place of manufacture and what chemometric methods are best suitable for matching. This particular investigation was only conducted on the HPIC data because (1) the data were multivariate and therefore suitable to variable selection and (2) a much larger set of samples were analyzed by HPIC than by IRMS or ICP-OES thereby permitting a more thorough investigation of classification performance using different combinations of variable selection and classification methods. Figure 2 shows resolution of NaCN and KCN based on four anions (unk5, SO4, OX, and PO4) into three main clusters (GE/BE, CZ, US/UK/ES) consistent with three factories (GE, CZ, and US) and inclusive of both KCN and NaCN samples. The four anions responsible for the observed clustering were previously selected using Fratio analysis of the 11 original anions and resulted in tighter clusters based on country of origin as opposed to using 11 anions.9 Given that the combined HPIC data in this paper had potentially new and significant sources of data variance (e.g., from different operators, different HPIC systems, and new stocks), HCA and PCA was performed on the combined HPIC data starting with all 11 anions (unk1, unk2, Cl, unk3, NO2, unk4, CO3, unk5, SO4, OX, and PO4) in order to detect any obvious outliers in regards to sample clustering according to source. That analysis resulted in the removal of carbonate and anion unk 2. Carbonate was removed because any relationship between it and cyanide source was masked by the relatively high and increasing background concentration from ambient carbon dioxide dissolving into the aqueous samples during sample preparation and analysis. Anion unk 2 was removed because it was present at abnormally high levels in the HPIC-2 samples. Apparently, contamination specific to anion unk 2 was introduced during HPIC-2, which was performed in a different laboratory with a different HPIC unit and operator than HPIC-1 and HPIC-3. Figures S-1 A, B and C in the Supporting Information do an excellent job of supporting the removal of anion unk 2. 12 ACS Paragon Plus Environment

Page 13 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

After removal of the carbonate and unk2 data, classification analysis was performed on the 27 cyanide stocks starting with the 9-anion (unk1, Cl, unk3, NO2, unk4, unk5, SO4, OX, PO4) profiles. The data set, which included all three batches of HPIC analysis, was divided into mutually exclusive training set and validation set pairs. Replicate anion profiles of the same sample were allowed to be either in the training or validation set, but never both simultaneously. Two such pairs of training and validation sets were developed for classification analysis to study performance under more realistic conditions. The first pair of training/validation sets (T/V-1) incorporated all KCN profiles in the training set, while all NaCN profiles were in the validation set. The second pair (T/V-2) for analysis included profiles from HPIC-1 and HPIC-2 in the training set with HPIC-3 profiles in the validation set. The purpose for T/V-1 was to evaluate classification modeling trained on a database of one type of cyanide stock (i.e., KCN) for its prediction ability over another type (i.e., NaCN). T/V-2 was chosen to evaluate the performance of prediction of cyanide stock samples analyzed using different operators and instruments, other than those that were used for training the models. Classification and variable (or feature) selection analysis were performed using tools from PLS-toolbox to analyze and investigate optimal performance. KNN, PLSDA, and SVMDA were utilized for classification analysis. Variable selection was performed using the F-ratio method, iPLS, and GAPLS. The performance of four anion variables (unk5, SO4, OX, and PO4) selected in reference 9 was also evaluated for comparison. The number of latent variables (LVs) required for PLS calibrations used in the models (PLSDA, iPLS, and GAPLS) were computed by the software using cross-validation. Cross-validation was performed using venetian blinds with 10 splits and an average 10% left-out data. KNN was performed for K=3. SVMDA classification was performed using c-SVM radial basis kernel. Table 2 and Table 3 show classification errors expressed in percentage for the two validation sets from T/V-1 and T/V-2, respectively, using different combinations of variable selection and classification methods. Classification error was calculated as the ratio of the number of misclassified profiles to the total number of profiles in the validation set; the interested reader may examine Table S4 and Table S5 to ascertain what specific sample profiles were misclassified. Classification errors for the two validation sets were as low as zero 13 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 25

for KNN and SVMDA while variable selection with iPLS (fwd) or F-ratio typically produced the lowest errors for those classification methods originally having errors greater than zero without variable selection. Classification errors of zero for T/V-1, suggest that unknown KCN samples can be reliably matched to a factory using known NaCN samples and vice versa. They also suggest at least one common anion source for KCN and NaCN made at the same factory. On the other hand, errors of zero for T/V-2 demonstrate the robustness of both anion impurities and the HPIC protocols given that the training samples were prepared and analyzed on different instruments approximately three to five years prior to the validation samples. Supervised Classification by Stock Classification of cyanide anion profiles according to stock for a given factory was also evaluated. HPIC profiles for four KCN stocks originating from US (K-US-A, K-US-C, K-US-G, and K-US-H) were examined using classification analysis. Each KCN stock (n=4) was represented by four separate solid aliquots separately dissolved in water and measured by HPIC analysis in triplicate. Specifically, there were two analytical replicates for each stock, complete with dissolution and dilution of each stock, for HPIC-1(on 8/2009) and for another HPIC batch performed on 5/2015 for a total of four aqueous samples for each stock (n=4) analyzed in triplicate; hence 48 anion profiles were generated and used in this investigation. Anion unk2 (as previously discussed) and PO4 (which was absent in the US stocks) were excluded from the analysis. Cross-validation errors were calculated to measure classification performance in the absence of a validation set. The results of classification are shown in Table 4 and Table 5. As shown in Table 4, classification based on four groups (K-US-A, K-US-C, K-US-G, and K-US-H) gave higher cross-validation errors for KNN, PLSDA, and SVMDA compared to three groups (Table 5) in which stocks C and H were treated as one group. The higher cross-validation errors obtained by treating stocks C and H separately indicate that their anion impurity profiles are not sufficiently distinguishable probably because parameters during their synthesis were relatively similar. Interestingly, both carbon and nitrogen stable isotope ratios for stocks C and H were statistically not distinguishable supporting the idea that stocks presumably created at different times at the same factory can have indistinguishable CAS. 14 ACS Paragon Plus Environment

Page 15 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

In regards to variable selection, it improved classification by stock for KNN and SVMDA and not PLSDA. For three-group classification, minimum errors from zero to 5.3% were obtained with the three classification methods demonstrating the potential of using anion impurities for matching a cyanide sample to a specific stock or group of stocks. Conclusion The past use of cyanides for nefarious purposes highlights the need for developing capabilities for the source attribution of cyanides and other chemical threat agents to support criminal investigation and prosecutions. This work demonstrated the potential for cyanide source attribution to a factory through an integration of anionic, elemental, and isotopic profiling. Six individual countries were listed for the origin of the 27 cyanide stocks; however, the supplier-provided information about country of origin was not completely supported by information about cyanide manufacturers, and the data were consistent with only three apparent sources for the cyanide stocks, i.e., one factory each in the Czech Republic, United States, and Germany. Additionally, IRMS carbon stable isotope ratios differentiated cyanides into two groups mostly consistent with two presumed facilities (one in the United States and the other in Germany), which demonstrated the potential of using a totally independent CAS (carbon isotope ratio) to corroborate the results from other CAS (anionic or elemental profiling). Chemometric analysis on the HPIC anion profiles from cyanides using various variable selection and classification methods supported five main conclusions. First, the matching or exclusion of a commercial cyanide sample to a factory using anion profiling appears achievable for solid cyanide stocks manufactured by respective facilities in the United States, Germany, and Czech Republic. The matching of a cyanide sample to a factory is independent of cyanide type (KCN or NaCN) suggesting at least one common anion source for KCN and NaCN made at that facility. Second, the matching of a commercial cyanide sample to a specific stock or stock group for a given plant (US was tested) is supported by analysis of anion impurities. This discovery may help independently corroborate stock matches obtained using IRMS carbon and nitrogen stable isotope ratios that have been previously demonstrated to be highly stock specific for cyanides.15 Third, the anion impurities in 15 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 25

cyanides and the implemented HPIC analysis protocols were robust given that the above mentioned outcomes were obtained using samples prepared and analyzed over a five year period by different HPIC instruments and operators. Fourth, variable selection reduced errors for those classification methods having errors greater than zero; iPLS (fwd) and then F-ratio typically providing the lowest errors. Lastly, KNN and SVMDA with variable selection had zero errors for factory classification for the two tested validation sets. Future work is merited to understand the limits and realistic error rates for the full integration of these and other chemical profiling techniques for sourcing cyanides to include a fundamental understanding of CAS origin. Ideally, this will require more cyanides from additional known sources and access to their reagents. Acknowledgements The authors would like to thank the following PNNL colleagues: Helen Kreuzer for providing the IRMS-1 data, Megan Nims for collection of the IRMS-2 data, Tom Farmer for collection of the ICP-OES data, and Ingrid Burgeson for collection of the HPIC-2 data. Funding for this work was provided by the Science and Technology Directorate, U.S. Department of Homeland Security under contract HSHQPM-11-X-00067.

Supporting Information available free of charge via the Internet at http://pubs.acs.org.

16 ACS Paragon Plus Environment

Page 17 of 25

Analytical Chemistry

Table 1. Stocks of commercial cyanides with the number of analyses performed and number of solid samples taken from each stock

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Stock IDa

Reported Country of Originb

Brandc

Product#

Lot#

Yeard

HPIC-1e 8/2009

K-US-A

USA

Columbus

4275AL

200800804

2008

6/2

K-US-C

USA

Sigma

207810

12830KH

2007

6/2

K-US-F

USA

Columbus

4275AL

200721214

2007

K-US-G

USA

Spectrum

P1280

VR0247

K-US-H

USA

Capitol

JTB-3080-01

G05423

HPIC-2e 9/2011

HPIC-3e 1/2015

IRMS-1e 11/2008

IRMS-2e 10/2014

ICPOES-1e 8/2009

2/2

6/2

9/9

6/2

6/2

2/2

6/2

2005

6/2

3/3

6/2

2008

6/2

2/2

6/2

K-US-6

USA

Alfa

L13273

USLF002608

2004*

K-GE-E

Germany

Spectrum

P1280

SN2860

2003

3/1

3/1

2/2

6/2

2/2

K-GE-2

Germany

Fluka

60178

1079981

2004*

3/1

2/2

K-GE-3

Germany

Fluka

60178

436248/1

2004*

3/1

2/2

K-GE-4

Germany

Riedel

31252

1110

2004*

3/1

2/2

3/1

6/2

K-BE-B

Belgium

Acros

38831

A0249685

2007

6/2

K-CZ-D

Czech R.

Sigma

60178

1350986

2007

6/2

K-ES-1

Spain

Acros

19660

A0291004

2009

N-US-1

USA

Sigma

380970

01207BH

2007

3/1

3/3

3/1

N-US-2

USA

Sigma

205222

79696MJ

2010*

3/1

4/4

3/1

N-US-3

USA

Sigma

431591

11005BB

2003

3/1

4/4

3/1

N-US-4

USA

Fisher

N/A

742568

1974

3/1

4/4

3/1

N-UK-1

U. Kingdom

Fluka

71430

440728/1

2004*

3/1

4/4

3/1

N-UK-2

U. Kingdom

Alfa

L13278

10149520

2010*

3/1

4/4

3/1

N-UK-3

U. Kingdom

Alfa

L13278

10173048

2012

N-GE-1

Germany

Fluka

71430

1346398

2007

3/1

N-GE-2

Germany

Fluka

71431

1297296

2006

3/1

2/2

N-GE-3

Germany

Fluka

71429

426688/1

2004*

3/1

4/4

4/4

3/1

N-BE-1

Belgium

Acros

42430

A0278135

2009

3/1

4/4

3/1

N-BE-2

Belgium

Acros

37031

A0200157

2008*

3/1

4/4

3/1

N-BE-3

Belgium

Acros

42430

A016873501

2004*

3/1

N-ES-1

Spain

Acros

42430

A0323184

2012

Number of analyses / number of solid samples =

3/1

ICPOES-2e 1/2011

19/19

6/2

17/17

6/2

3/1

4/4

3/1

3/1

4/4

3/1

3/3

42/14

3/1

3/3 3/1

48/16

3/1

30/10

3/1 4/4

73/73

50/50

48/16

36/12

a

K = KCN; N = NaCN b Country of origin provided by supplier. c Full brand names: Acros Organics; Alfa Aesar; Capitol Scientific; Columbus Chemical Industries; Fisher Scientific; Fluka Analytical; Riedel-de Haën; Sigma-Aldrich; Spectrum Chemical d Year stock was manufactured according to supplier or year when quality control analysis was performed by supplier unless marked by “*” which indicates year received by laboratory. e Batch analysis name and date plus number of analyses / number of solid samples for each stock for a given analysis batch, e.g., “3/1” means 3 analyses were performed for one solid sample; “6/2” means 3 analyses per solid sample were performed for 2 solid samples; “3/3” means one analysis per solid sample was performed for 3 solid samples.

17 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 18 of 25

Table 2. Three groupa classification errors for T/V-1b Validation-41 profiles Variablesc Anions % error (# misclassified) KNN PLSDA SVMDA 2.4 (1) 0 (0) 4.9 (2) All-9 9 0 (0) 27 (11) 2.4 (1) Previous-4 4 2.4 (1) 32 (13) 2.4 (1) F-ratio 3 0 (0) 4.9 (2) 0 (0) iPLS fwd. 3 34 (14) 34 (14) 22 (9) iPLS rev. 7 9.8 (4) 9.8 (4) 12 (5) GAPLS 3 a

Groups: (1) GE/BE, (2) CZ, (3) US/UK/ES Training (T) set: KCN from GE (n=15), BE (n=12), CZ (n=6), US (n=33), ES (n=3) Validation (V) set: NaCN from GE (n=6), BE (n=9), US (n=11), UK (n=12), ES (n=3) c All-9: unk1, Cl, unk3, NO2, unk4, unk5, SO4, OX, PO4 Previous-4: unk5, SO4, OX, PO4 F-ratio: unk1, SO4, PO4 iPLS(fwd): unk4, unk5, PO4 iPLS(rev): unk3, NO2, unk4, unk5, SO4, OX, PO4 GAPLS: unk1, OX, PO4 b

Table 3. Three groupa classification errors for T/V-2b Validation-30 profiles Variablesd Anions % error (# misclassified) KNN PLSDA SVMDA 6.7 (2) 20 (6) 0 (0) All-9 9 10 (3) 10 (3) 27 (8) Previous-4 4 0 (0) 20 (6) 0 (0) F-ratio 6 0 (0) 20 (6) 0 (0) iPLS fwd. 2 13 (4) 30 (9) 0 (0) iPLS rev. 7 27 (8) 13 (4) 10 (3) GAPLS 6 a

Groups: (1) GE/BE, (2) CZ, (3) US/UK/ES Training (T) set: KCN and NaCN from HPIC-1 and HPIC-2 for GE (n=12), BE (n=15), CZ (n=6), US (n=41), UK (n=6) Validation (V) set: KCN and NaCN from HPIC-3 for GE (n=12), BE (n=3), US (n=3), UK (n=6), ES (n=6) c All-9: unk1, Cl, unk3, NO2, unk4, unk5, SO4, OX, PO4 Previous-4: unk5, SO4, OX, PO4 F-ratio: unk1, unk3,NO2, unk5, SO4, PO4 iPLS(fwd): unk5, PO4 iPLS(rev): Cl, unk3, NO2, unk5, SO4, OX, PO4 GAPLS: unk3, unk4, unk5, SO4, OX, PO4 b

18 ACS Paragon Plus Environment

Page 19 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Table 4. Four stock group classification errors for KCN stocks: (1) A, (2) C, (3) G, (4) H Crossvalidation-48 profiles Variablesa Anions % error KNN PLSDA SVMDA 13 21 9.7 All-8 8 8.3 32 9.7 F-ratio 5 14 26 21 iPLS fwd. 7 15 25 6.9 iPLS rev. 6 GAPLS 4 9.7 33 11 a

All-8: F-ratio: iPLS(fwd): iPLS(rev): GAPLS-4:

unk1, Cl, unk3, NO2, unk4, unk5, SO4, OX Cl, unk3, NO2, unk4, SO4 unk1, unk3, NO2, unk4, unk5, SO4, OX unk1, unk3, unk4, unk5, SO4, OX Cl, unk3, NO2, SO4

Table 5. Three stock group classification errors for KCN stocks: (1) A, (2) C/H, (3) G Crossvalidation-48 profiles Variablesa Anions % error KNN PLSDA SVMDA 3.9 5.3 2.1 All-8 8 5.3 27 5.1 F-ratio 4 4.4 34 3.2 iPLS fwd. 2 3.2 17 1.9 iPLS rev. 6 GAPLS 5 2.1 21 0 a

All-8: F-ratio: iPLS(fwd): iPLS(rev): GAPLS-4:

unk1, Cl, unk3, NO2, unk4, unk5, SO4, OX unk3, NO2, unk4, SO4 NO2, OX unk3, NO2, unk4, unk5, SO4, OX unk3, NO2, unk4, unk5, SO4

19 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 20 of 25

References (1) Khan, A. S.; Levitt, A. M.; Saje, M. J. Morb. Mortal. Wkly. Rep. 2000, 49, 1-14. (2) Keim, M. E. Prehospital and Disaster Medicine 2006, 21, s56-s60. (3) Shear, M. D. Letter Bound For White House Tests Positive for Cyanide. The New York Times, 2015. (4) Monterey Institute of International Studies. Chronology of Aum Shinrikyo's CBW Activities. 2001. Avaliable at http://cns.miis.edu/reports/pdfs/aum_chrn.pdf. (5) Sharp, A. G. The Chemistry of Cyano Complexes of the Transition Metals; Academic Press: London, UK, 1976. (6) Vogt, D.; Vogt, J. G. Biochemistry; John Wiley and Sons: New York, NY, 2004. (7) Rubo, A.; Kellens, R.; Reddy, J.; Steier, N.; Hasenpusch, W. Alkali Metal Cyandies: Ullmann's Encyclopedia of Idustrial Chemistry; Wiley-VCH: Verlag, GmbH, 2006. (8) Favela, K. H.; Bohmann, J. A.; Williamson, W. S. Forensic Sci. Int. 2012, 217, 39-49. (9) Fraga, C. G.; Farmer, O. T.; Carman, A. J. Talanta 2010, 83, 1166-1172. (10) Fraga, C. G.; Perez-Acosta, G. A.; Crenshaw, M. D.; Wallace, K.; Mong, G. M.; Colburn, H. A. Anal. Chem. 2011, 83, 9564-9572. (11) Fraga, C. G.; Wahl, J. H.; Nuñez, S. P. Forensic Sci. Int. 2011, 210, 164-169. (12) Hoggard, J. C.; Wahl, J. H.; Synovec, R. E.; Mong, G. M.; Fraga, C. G. Anal. Chem. 2010, 82, 689-698. (13) Mazzitelli, C. L.; Re, M. A.; Reaves, M. A.; Acevedo, C. A.; Straight, S. D.; Chipuk, J. E. Anal. Chem. 2012, 84, 6661-6671. (14) Tea, I.; Antheaume, I.; Zhang, B.-L. Forensic Sci.Int. 2012, 217, 168-173. (15) Kreuzer, H. W.; Horita, J.; Moran, J. J.; Tomkins, B. A.; Janszen, D. B.; Carman, A. J. J. Forensic Sci. 2011, 57, 75-79. (16) Brust, H.; Koeberg, M.; Heijden, A. v. d.; Wiarda, W.; Mugler, I.; Schrader, M.; Vivo-Truyols, G.; Schoenmakers, P.; Asten, A. v. Forensic Sci. Int. 2015, 248, 101-112. (17) Barker, M.; Rayens, W. J. Chemometrics 2003, 17, 166-173. (18) Ballabio, D.; Consonni, V. Anal. Meth.2013, 5, 3790-3798. (19) IHS Chemicals. 2006 Chemical Economic Handbook: Sodium Cyanide, Available at https://www.ihs.com/products/chemical-economics-handbooks.html. (20) IHS Chemicals. 2010 Directory of Chemical Producers, Available at https://www.ihs.com/products/chemical-companies-producers.html. (21) Rules of Origin, Title 19, Part 102, Code of Federal Regulations. Available at http://www.gpo.gov/fdsys/granule/CFR-2012-title19-vol1/CFR-2012-title19-vol1-part102/content-detail.html. (22) Manager, Cyanide Producer. Personal communication, May 2015.

20 ACS Paragon Plus Environment

Page 21 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Figures

Figure 1. Anion chromatograms for two aqueous cyanide samples prepared from KCN stock K-GE-2 and NaCN stock N-GE-1. The reported anion concentrations are in parts-per-million (ppm) (w/w) in solid cyanide. The peak for unk2 in the lower plot is cut-off to emphasize the smaller peaks. The CN- anion is not detectable.

21 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 22 of 25

Figure 2. HCA dendrogram generated using Ward’s method on the area-normalized and autoscaled combined HPIC data set (HPIC-1, HPIC-2, and HPIC-3) consisting of 120 cyanide sample profiles with peak areas for four anion impurities (unk5, SO4, OX, and PO4). KCN and NaCN samples from all three sets clustered into three groups designated by countries of origins: (1) US/UK/ES, (2) GE/BE and (3) CZ. Apparent outlier samples from K-US-6 cluster with GE/BE group and samples from N-GE-3 and N-BE-3 cluster with the US/UK/ES group rather than the group corresponding to their reported country of origins.

22 ACS Paragon Plus Environment

Page 23 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Figure 3. HCA dendrogram for the combined IRMS data set (IRMS-1 and IRMS-2) consisting of averaged stable carbon isotope ratios (δ13C) for 27 cyanide stocks. KCN and NaCN stocks cluster into main two groups: (1) US/UK/ES and (2) GE/BE. Two stock samples reportedly from US (K-US-6 and N-US-4) cluster with GE/BE group. Stock samples reportedly from Germany (N-GE-3), Belgium (N-BE-3) and Czech Republic (KCZ-D) cluster with US/UK/ES group.

23 ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 24 of 25

Figure 4. HCA dendrogram generated using Ward’s method on area-normalized and autoscaled combined ICPOES data (ICP-OES-1 and ICP-OES-2) consisting of 84 cyanide elemental profiles with concentrations for 9 elements (Ba, Ca, Fe, Mg, P, Rb, S, Si, and Sr) from 20 cyanide stocks. KCN and NaCN samples cluster into three main groups: (1) US/UK, (2) GE/BE, and (3)CZ. Samples of two stocks reportedly from Germany (N-GE3) and Belgium (N-BE-3) clustered with the US/UK group.

24 ACS Paragon Plus Environment

Page 25 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

TOC Only

25 ACS Paragon Plus Environment